Bulk catalog data

Every public model in one daily file: files, sizes, SHA-256, availability, license label and download URL.

One download a day gives you every public catalog row. Use it instead of polling model pages or /api/manifests/<id>.

Latest snapshot: 2026-11-09, 177641 rows. JSONL.gz 6.8 MB, Parquet 48.3 MB.

Download JSONL.gz (one model per line)
Download Parquet
Snapshot details (JSON): date, row count, SHA-256
Changes since a date: https://huggingbay.xyz/api/v1/catalog/changes?since=YYYY-MM-DD

What each row has

id: Catalog id (use it with /artifact/<id> and the API).
repo: Repository name.
license: License label as recorded (labels only). An 'unknown' label does not block a download: a file is downloadable when its row has hosted: true and a downloadUrl.
availability: verdict, downloadReady, hostedFileCount, checkedAt and source from the same resolver the site uses at build time.
files: path, sizeBytes, sha256 (recorded), hosted, downloadUrl (our /api/downloads route; it re-checks on every request). Up to 500 per row; filesTruncated marks more.
runnableOn: Smallest weight set and minRamGb using the Finder size budget (file size <= 0.6 x RAM). A size hint, not a run test.

Try it

curl -sL https://huggingbay.xyz/api/v1/catalog/snapshot.jsonl.gz | gunzip | head -n 1
curl -s "https://huggingbay.xyz/api/v1/catalog/changes?since=2026-11-08" | head -n 3

Changes are kept for the last 3 days, back to 2026-11-06. Older than that, download the full snapshot.

Rebuilt once a day (UTC). Dated files never change; undated URLs redirect to the newest day. Free tier: 10 file downloads per hour per IP on each server instance; API keys get 120. Redirects, HEAD and 304 answers are free. One download a day is all you need. Need more? Request crawl access.